Blanchard-Kahn conditions: sensitivity to parameter values
Posted: Thu Feb 02, 2012 11:51 amHi all,
I know that there are many posts in the forum concerning the BK conditions, and I have looked through all of them before asking this question. I have a very big model (72 equations) and the model sometimes satisfies the conditions and sometimes not, depending on parameter values which are in most cases logical. Note that:
1) I have solved the steady-state analytically and log-linearised the model by hand. I am sure the model is well specified.
2) I checked my code over and over again, I am pretty sure it is correct and that the timing of the variables is right.
3) When the conditions fail, they do so because of one eigenvalue; i.e. the reason for failure is always "There are 7 eigenvalue(s) larger than 1 in modulus
for 8 forward-looking variable(s)".
4) Normally, I have GHH preferences. For all the cases that the BK conditions fail with these type of preferences, the model runs normally when I use a simple log-utility.
Considering all the above, can I say with confidence that the problem is technical, in the sense that due to its dimensionality the model exhibits this sensitivity to the parameter values? If that is the case, would decreasing the dimensionality help? How could I perform sensitivity analysis when I face this issue?
Any other advise/ hint would be greatly appreciated. Thanks a lot.
Kyriacos
I know that there are many posts in the forum concerning the BK conditions, and I have looked through all of them before asking this question. I have a very big model (72 equations) and the model sometimes satisfies the conditions and sometimes not, depending on parameter values which are in most cases logical. Note that:
1) I have solved the steady-state analytically and log-linearised the model by hand. I am sure the model is well specified.
2) I checked my code over and over again, I am pretty sure it is correct and that the timing of the variables is right.
3) When the conditions fail, they do so because of one eigenvalue; i.e. the reason for failure is always "There are 7 eigenvalue(s) larger than 1 in modulus
for 8 forward-looking variable(s)".
4) Normally, I have GHH preferences. For all the cases that the BK conditions fail with these type of preferences, the model runs normally when I use a simple log-utility.
Considering all the above, can I say with confidence that the problem is technical, in the sense that due to its dimensionality the model exhibits this sensitivity to the parameter values? If that is the case, would decreasing the dimensionality help? How could I perform sensitivity analysis when I face this issue?
Any other advise/ hint would be greatly appreciated. Thanks a lot.
Kyriacos